def drain_nodes(node_names):
"""
Drain nodes
Args:
node_names (list): The names of the nodes
Raises:
TimeoutExpired: in case drain command fails to complete in time
"""
ocp = OCP(kind="node")
node_names_str = " ".join(node_names)
log.info(f"Draining nodes {node_names_str}")
try:
ocp.exec_oc_cmd(
f"adm drain {node_names_str} --force=true --ignore-daemonsets "
f"--delete-local-data",
timeout=1800,
)
except TimeoutExpired:
ct_pod = pod.get_ceph_tools_pod()
ceph_status = ct_pod.exec_cmd_on_pod("ceph status",
out_yaml_format=False)
log.error(
f"Drain command failed to complete. Ceph status: {ceph_status}")
# TODO: Add re-balance status once pull/1679 is merged
raise
def schedule_nodes(node_names):
"""
Change nodes to be scheduled
Args:
node_names (list): The names of the nodes
"""
ocp = OCP(kind="node")
node_names_str = " ".join(node_names)
ocp.exec_oc_cmd(f"adm uncordon {node_names_str}")
log.info(f"Scheduling nodes {node_names_str}")
wait_for_nodes_status(node_names)
def drain_nodes(node_names):
"""
Drain nodes
Args:
node_names (list): The names of the nodes
"""
ocp = OCP(kind='node')
node_names_str = ' '.join(node_names)
log.info(f'Draining nodes {node_names_str}')
ocp.exec_oc_cmd(
f"adm drain {node_names_str} --force=true --ignore-daemonsets "
f"--delete-local-data")
def unschedule_nodes(node_names):
"""
Change nodes to be unscheduled
Args:
node_names (list): The names of the nodes
"""
ocp = OCP(kind="node")
node_names_str = " ".join(node_names)
log.info(f"Unscheduling nodes {node_names_str}")
ocp.exec_oc_cmd(f"adm cordon {node_names_str}")
wait_for_nodes_status(node_names, status=constants.NODE_READY_SCHEDULING_DISABLED)
def scale_down_deployments(node_name):
"""
Scale down the deployments of a node as described in the documents
of node replacement with LSO
Args:
node_name (str): The node name
"""
ocp = OCP(kind="node", namespace=defaults.ROOK_CLUSTER_NAMESPACE)
pods_to_scale_down = get_node_pods_to_scale_down(node_name)
for p in pods_to_scale_down:
deployment_name = pod.get_deployment_name(p.name)
log.info(f"Scale down deploymet {deployment_name}")
ocp.exec_oc_cmd(f"scale deployment {deployment_name} --replicas=0")
log.info("Scale down rook-ceph-crashcollector")
ocp.exec_oc_cmd(
f"scale deployment --selector=app=rook-ceph-crashcollector,"
f"node_name='{node_name}' --replicas=0")
def taint_ocs_nodes(nodes_to_taint=None):
"""
Function to taint nodes with "node.ocs.openshift.io/storage=true:NoSchedule"
Args:
nodes_to_taint (list): Nodes to taint
"""
if not check_taint_on_ocs_nodes():
ocp = OCP()
ocs_nodes = get_ocs_nodes()
nodes_to_taint = nodes_to_taint if nodes_to_taint else ocs_nodes
log.info(f"Taint nodes with taint: " f"{constants.OPERATOR_NODE_TAINT}")
for node in nodes_to_taint:
taint_cmd = f"adm taint nodes {node.name} {constants.OPERATOR_NODE_TAINT}"
ocp.exec_oc_cmd(command=taint_cmd)
else:
log.info(
f"One or more nodes already have taint {constants.OPERATOR_NODE_TAINT} "
)
def remove_nodes(nodes):
"""
Remove the nodes from cluster
Args:
nodes (list): list of node instances to remove from cluster
"""
ocp = OCP(kind="node")
node_names = [node.get().get("metadata").get("name") for node in nodes]
node_names_str = " ".join(node_names)
# unschedule node
unschedule_nodes(node_names)
# Drain all the pods from the node
drain_nodes(node_names)
# delete the nodes
log.info(f"Deleting nodes {node_names_str}")
ocp.exec_oc_cmd(f"delete nodes {node_names_str}")
def untaint_ocs_nodes(taint=constants.OPERATOR_NODE_TAINT, nodes_to_untaint=None):
"""
Function to remove taints from nodes
Args:
taint (str): taint to use
nodes_to_taint (list): list of nodes to untaint
Return:
bool: True if untainted, false otherwise
"""
if check_taint_on_nodes():
ocp = OCP()
ocs_nodes = get_ocs_nodes()
nodes_to_taint = nodes_to_untaint if nodes_to_untaint else ocs_nodes
for node in nodes_to_taint:
taint_cmd = f"adm taint nodes {node.name} {taint}-"
ocp.exec_oc_cmd(command=taint_cmd)
log.info(f"Untainted {node.name}")
return True
return False
def add_scc_policy(sa_name, namespace):
"""
Adding ServiceAccount to scc privileged
Args:
sa_name (str): ServiceAccount name
namespace (str): The namespace for the scc_policy creation
"""
ocp = OCP()
out = ocp.exec_oc_cmd(
command=
f"adm policy add-scc-to-user privileged system:serviceaccount:{namespace}:{sa_name}",
out_yaml_format=False)
logger.info(out)
def remove_scc_policy(sa_name, namespace):
"""
Removing ServiceAccount from scc privileged
Args:
sa_name (str): ServiceAccount name
namespace (str): The namespace for the scc_policy deletion
"""
ocp = OCP()
out = ocp.exec_oc_cmd(
command=
f"adm policy remove-scc-from-user privileged system:serviceaccount:{namespace}:{sa_name}",
out_yaml_format=False)
logger.info(out)
